List your company
ResearchFormerly WizWand

Multimodal Model Evaluation on MME (MME Metric)

2,489.7MME ScoreInternVL3.5-38B
1,092.83441,455.48221,818.132,180.7778Feb 17, 2025May 7, 2025Jul 25, 2025Oct 13, 2025Dec 31, 2025Mar 20, 2026Jun 8, 2026
Updated 4mo ago

Evaluation results

MethodLinks
InternVL3.5-38BRatio=100%, Backbone=InternVL3.5-38B2026.052,489.7
FP16Spiking=false, Time Step=N/A, Backbone=Qwen2VL-72B2026.042,480
SpikeMLLM (QuaRot+MSTS+TC-LIF)Spiking=true, Time Step=3/4, Backbone=Qwen2VL-72B2026.042,451
F3ARatio=60%, Backbone=InternVL3.5-38B2026.052,430.9
FULLModel=Qwen3-VL (4B), Selection Ratio=100%2026.052,429.7
FastVRatio=60%, Backbone=InternVL3.5-38B2026.052,426.6
SpikeMLLM(SQ)Spiking=true, Time Step=255, Backbone=Qwen2VL-72B2026.042,419
CDPrunerRatio=60%, Backbone=InternVL3.5-38B2026.052,417.8
VisionZipRatio=60%, Backbone=InternVL3.5-38B2026.052,403.6
RANDOMModel=Qwen3-VL (4B), Selection Ratio=20%2026.052,401
Honeybee-Remake-SEED-200KModel=Qwen3-VL (4B), Selection Ratio=20%2026.052,398.1
DivPruneRatio=60%, Backbone=InternVL3.5-38B2026.052,389.7
F3ARatio=40%, Backbone=InternVL3.5-38B2026.052,365.3
Training-Free Debiasing Inference StrategyModel=Qwen2.5-VL-7B, Strategy=Ours2026.052,355.2
DMASModel=Qwen2.5-VL-7B, Strategy=DMAS2026.052,346
SIDModel=Qwen2.5-VL-7B, Strategy=SID2026.052,342.8
VisionZipRatio=40%, Backbone=InternVL3.5-38B2026.052,341.8
VISTAModel=Qwen2.5-VL-7B, Strategy=VISTA2026.052,338.1
VCDModel=Qwen2.5-VL-7B, Strategy=VCD2026.052,335.7
SpikeMLLM(MBQ)Spiking=true, Time Step=255, Backbone=Qwen2VL-72B2026.042,331
Zero-shotModel=Qwen2.5-VL-7B, Strategy=Zero-shot2026.052,328.1
CDPrunerRatio=40%, Backbone=InternVL3.5-38B2026.052,321.4
Vanilla2026.042,319
DivPruneRatio=40%, Backbone=InternVL3.5-38B2026.052,318.2
BASEModel=Qwen3-VL (4B), Selection Ratio=0%2026.052,279.5
Honeybee-Remake-SEED-200KModel=LLaVA-OneVision-1.5 (4B), Selection Ratio=20%2026.052,260.8
RANDOMModel=LLaVA-OneVision-1.5 (4B), Selection Ratio=20%2026.052,256.7
BASEModel=LLaVA-OneVision-1.5 (4B), Selection Ratio=0%2026.052,248.7
FULLModel=LLaVA-OneVision-1.5 (4B), Selection Ratio=100%2026.052,244.7
FastVRatio=40%, Backbone=InternVL3.5-38B2026.052,235.6
F3ARatio=20%, Backbone=InternVL3.5-38B2026.052,230.7
FastVRatio=20%, Backbone=InternVL3.5-38B2026.052,158.3
VisionZipRatio=20%, Backbone=InternVL3.5-38B2026.052,146.8
DASHViT retain tokens=50%, LLM retain tokens=50%2026.042,116
DART (ViT)ViT retain tokens=50%, LLM retain tokens=50%2026.042,113
DivPruneRatio=20%, Backbone=InternVL3.5-38B2026.052,097.1
CDPrunerRatio=20%, Backbone=InternVL3.5-38B2026.052,089.6
ToMeViT retain tokens=50%, LLM retain tokens=50%2026.042,007
DASHViT retain tokens=35%, LLM retain tokens=35%2026.041,987
DART (ViT)ViT retain tokens=35%, LLM retain tokens=35%2026.041,968
LLaVA-Qwen-7B (Instruct)Model=LLaVA-Qwen-7B (Instruct)2025.021,935.4
ToFuViT retain tokens=50%, LLM retain tokens=50%2026.041,933
LLaVA-Qwen-7B (Base)Model=LLaVA-Qwen-7B (Base)2025.021,923.1
PRISM-Qwen-7B (Instruct)Model=PRISM-Qwen-7B (Instruct)2025.021,908.6
PRISM-Qwen-7B (Base)Model=PRISM-Qwen-7B (Base)2025.021,895.8
LLaVA-Llama-8BModel=LLaVA-Llama-8B2025.021,889.7
PRISM-Llama-8BModel=PRISM-Llama-8B2025.021,862.5
PRISM-13BModel=PRISM-13B2025.021,846
ToMeViT retain tokens=35%, LLM retain tokens=35%2026.041,842
ToFuViT retain tokens=35%, LLM retain tokens=35%2026.041,839
LLaVA-Vicuna-13BModel=LLaVA-Vicuna-13B2025.021,826.7
LLaVA-Vicuna-7BModel=LLaVA-Vicuna-7B2025.021,822.6
PRISM-7BModel=PRISM-7B2025.021,800
SyVLA2026.061,795
PRISM-3BModel=PRISM-3B2025.021,790.5
LLaVA-1.5-13BToken=576, Backbone=LLaVA-1.5-13B2026.041,789.68
RCPToken=192, Backbone=LLaVA-1.5-13B2026.041,786.1
Training-Free Debiasing Inference StrategyModel=LLaVA-1.5-7B, Strategy=Ours2026.051,771.3
DMASModel=LLaVA-1.5-7B, Strategy=DMAS2026.051,770.4
SIDModel=LLaVA-1.5-7B, Strategy=SID2026.051,769.2
VCDModel=LLaVA-1.5-7B, Strategy=VCD2026.051,766
LLaVA-Phi2-3BModel=LLaVA-Phi2-3B2025.021,765.7
Zero-shotModel=LLaVA-1.5-7B, Strategy=Zero-shot2026.051,762.5
VISTAModel=LLaVA-1.5-7B, Strategy=VISTA2026.051,760
RCPToken=128, Backbone=LLaVA-1.5-13B2026.041,742.43
RCPToken=64, Backbone=LLaVA-1.5-13B2026.041,660.29
MAGICTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.051,551.5
Full-FinetuneTarget Model=LLaVA-1.5-13B, Data Budget=100%2026.051,541.7
EL2NTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.051,531
COINCIDETarget Model=LLaVA-1.5-13B, Data Budget=20%2026.051,516.8
FastVToken=192, Backbone=LLaVA-1.5-13B2026.041,515.86
PerplexityTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.051,508.8
RandomTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.051,484.9
Self-SupTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.051,463.8
D2-PruningTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.051,463
Self-FilterTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.051,446.9
ChatVLA2026.061,435
CLIP-ScoreTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.051,426.3
SemDedupTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.051,397.6
Wall-Oss2026.061,146.56